Research on Real-Time Face Recognition and Tracking Model Based on Embedded Platform

Human face recognition and tracking technology is widely applied in various fields. However, those algorithms based on deep learning run on resource-constrained embedded devices remains a significant challenge. A multi-algorithm fusion human face recognition and tracking model is proposed for the application on a Raspberry Pi–controlled intelligent robot in this paper. The model composes with MTCNN algorithm, SORT algorithm and Inception-ResNet algorithm. MTCNN provides the initial face detection results by transfer learning; SORT assigns identities and tracks detected faces, maintaining stability and accuracy even under unideal environments; and Inception-ResNet performs face recognition by computing face feature similarity across individuals. The proposed model achieves over 99% accuracy on LFW dataset by the optimized similarity threshold. Lightweight techniques, such as multi-threading and code structure refinement, are adopted to reduce memory consumption and enhance real-time performance for the application of the proposed model on the Raspberry Pi–controlled intelligent robot. Test results show that the proposed model runs stably on the Raspberry Pi platform, satisfies real-time processing requirements, and provides a practical reference for human face recognition and tracking algorithms on embedded robot platforms.